Scaffold
A deterministic desktop prompt compiler. It turns a vague request into a structured, verifiable prompt for any AI system — and it doesn't call an AI to do it.
Most prompt tools guess at what you meant and hope. Scaffold makes it say what it doesn't know, out loud, before you find out the hard way.
What it does
You type "make me a horror game." Scaffold does the rest — deterministically.
No AI in the loop
Scaffold's own reasoning is rule-based and local. No OpenAI, no Anthropic, no Gemini call happens anywhere inside it — the compiler itself never phones out.
Shows its work
Every inferred requirement, assumption, and architecture note is traceable to which specialist produced it and why — never a silent black-box rewrite of your prompt.
Asks instead of guessing
Genuinely ambiguous gaps get surfaced as real questions you answer, not quietly invented defaults you'd only discover later.
This is relatedness matching against fields you've already answered, not full semantic conflict detection across the whole spec — a recommendation Scaffold hasn't connected to anything yet still gets surfaced normally.
How it works
Seven specialist reviewers, one compiled result.
Intent, domain, and requirements — parsed, not guessed
A real offline NLP layer reads grammar, negation, and synonyms to figure out what you're actually asking for, before anything gets classified.
Seven specialists — Architect, Technical, UX, Security, Creative, QA, Constraint
Each one is a deterministic rule engine with its own job, not a generic pass. Their output is critiqued and cross-checked for contradictions before anything is finalized.
Three modes — Quick, Architect, Master
Quick for a fast pass, Architect for a full structured spec, Master for genuine multi-round deliberation when it's worth the extra time. You choose the depth.
Local-first, one SINVAUX account.
Scaffold runs entirely on your machine — your prompts and history never leave it. The same SINVAUX account you'd use across our other products signs you in here too, once it's ready for wider testing.